Security, Compliance, and Data Protection

The Question Every CIO Asks

Before authorizing ChatGPT within the enterprise, the IT department almost always asks the same question: where do our data go? A vague answer sends the file straight to the “review later” pile. A factual answer, backed by contractual commitments and verifiable audits, unlocks approval. This lesson gives you what you need to reassure decision-makers when justified—and to alert them when necessary.

OpenAI’s Data Model for Enterprises

On Business, Enterprise, and Edu plans, OpenAI commits to not training its models on your data by default, covering both inputs and outputs. You retain ownership of these inputs and outputs to the extent permitted by law. These commitments appear in OpenAI’s business privacy pages and contractual terms: these are the documents your legal team will want to rely on.

Data is encrypted at rest using AES-256 and in transit using TLS 1.2 or higher. Retention depends on the plan. Conversations remain saved until deleted, and deletion becomes permanent within 30 days; on Enterprise and Edu, a workspace retention policy can erase them sooner, and this also applies to saved files. A temporary chat does not appear in history, but a copy may be retained for up to 30 days for security reasons.

Comparing this to individual plans highlights the value of this commitment. On Free, Plus, or Pro, conversations may be used for training unless the user disables the option in their data controls, and even then, thumbs-up or thumbs-down feedback on a response can authorize the use of the entire conversation. On business plans, this is not an individual setting, and feedback on a response does not authorize training on the conversation.

Compliance and Certifications

OpenAI states it has passed a SOC 2 audit confirming that its controls meet security and privacy standards. Reports and attestations are available via its Trust Portal: this is typically the document a CISO requests before signing.

From a GDPR perspective, OpenAI acts as a data processor for your workspace data. Its Data Processing Addendum (DPA) complements and is incorporated into the OpenAI Services Agreement. Transfers outside the European Economic Area and Switzerland rely on Standard Contractual Clauses or an adequacy decision. New Enterprise and Edu customers can store their content in Europe (EEA and Switzerland) and run inference there as well. On Business, the choice of storage region is rolling out; it does not cover inference, and abuse prevention logs, including a temporary copy of exchanges, remain in the United States.

More specific needs require a contracted offer: the BAA required by HIPAA in the United States, zero-data retention, or customer-managed encryption keys (EKM) are not available via self-service.

What to Verify Before Deploying

Granting access requires a series of preliminary checks, and the order in which you conduct them matters. First, verify that your plan explicitly excludes data from training: this is the default on Business, Enterprise, and Edu, but read it in the applicable terms rather than relying on a marketing page. Next, review the DPA integrated into the contract with your Data Protection Officer.

On Enterprise or Edu, set the retention policy according to your internal rules. Document the purposes of processing: your GDPR record of processing activities must include ChatGPT just like your other tools, and omissions are immediately visible during an audit. Finally, inform your employees about how their data is used within the tool.

Gray Areas No Setting Can Solve

The first concerns sensitive data entered in prompts. Even if excluded from training, this data transits through OpenAI’s servers, which requires explicit rules on what can and cannot be entered: customer personal data, non-public financial information, and trade secrets.

The second involves memory, which retains information from one conversation to the next. On Enterprise, administrators configure this by role, including shared memory and enhanced memory; however, you must make this decision rather than letting the default setting decide for you.

The third concerns shared files. As soon as you connect Google Drive or SharePoint, ChatGPT accesses documents the connected account can open, and a folder open to the entire company remains visible from ChatGPT. Remember that the greatest security risk is not a flaw in OpenAI’s encryption, but a rushed employee pasting a confidential contract into the free version from a personal phone due to lack of professional access. Providing managed access costs less than letting teams fend for themselves with accounts over which you have no visibility.

Your Turn

Prepare your CIO’s dossier. Gather on a single page the target plan, the training commitment, encryption details, applicable retention, storage region, the DPA, and the reference for the SOC 2 report to request via the Trust Portal. Have your Data Protection Officer review it before the approval meeting.

Official sources: Enterprise privacy at OpenAI · Business data privacy, security, and compliance · OpenAI Data Processing Addendum (OpenAI) · Chat and file retention in ChatGPT · Data residency and inference residency for ChatGPT · Where your ChatGPT Business content is stored · How your data is used to improve model performance (OpenAI Help Center)

Key Takeaways

• On Business, Enterprise, and Edu, your data is not used for training by default, and feedback on a response does not change this.
• AES-256 encryption at rest and TLS 1.2 or higher in transit; permanent deletion within 30 days, configurable retention on Enterprise and Edu.
• OpenAI has passed a SOC 2 audit; its DPA, integrated into the contract, governs transfers via Standard Contractual Clauses.
• European residency: storage and inference on Enterprise and Edu; storage only and rolling out on Business.
• Residual risks are human: sensitive data in prompts, unmanaged memory, and misconfigured sharing permissions.